Title: NHL Advanced Stats: 2012-13 Season Primer
Date: January 11, 2013
Original Source: Nucks Misconduct
Synopsis: I will now be writing for Nucks Misconduct, a Vancouver Canucks blog on the SBNation network. My debut article was an advanced statistics primer.
I’m going to be dropping in here at Nucks Misconduct every so often, usually talking stats. I know Advanced Stats aren’t for everyone, especially in the hockey world where they’ve been a little slower to develop/disseminate than in baseball and even basketball.
Hockey is a very complex game with little in the way of “results” (goals), so it can be difficult to quantify and measure. Stats will never outweigh watching the games and developing qualitative opinion, I don’t think, but they can certainly be helpful in affirming what we think we see and maybe even pointing us in the right direction when we’re stumped.
As an example, imagine two players:
Harold Hustle plays centre on the third line, wins a lot of faceoffs, plays with two low-skill, checking-only players and almost always faces off against an opponent’s top scorers. In 82 games, he scores 10 goals with 15 assists (25 points) with a +4 rating in 13 minutes of ice time a night.
Doug Dangles plays centre on the first line, wins half his faceoffs, plays with a pair of snipers on either side and rarely matches up against another team’s top line. In 82 games, he scores 28 goals with 55 assists (83 points) and a +8 rating in 19 minutes of ice time a night.
Now, if you were to just take the hockey cards for these two players, you’d certainly conclude Dangles is the much better player. And you probably wouldn’t be wrong – there’s a reason he gets the ice time, line-up position and role that he has.However, without watching the games and really analyzing the two players, simply saying Dangles was 58 points better than Hustle is probably incorrect.
I’m not going to dive too deeply into the math behind the stats (at least not today).
Instead, I’ll just break down some of the stats that I’ll likely reference from time to time, either in stat-specific pieces or in more general articles. I’ll also link to the best place to find said stats.
Per-60 Minutes – a lot of the time I will translate stats to a per-60-minute basis. This is simply to standardized scoring, penalties, etc on a time-balanced scale so that time on ice doesn’t wildly effect our analysis. For example, while Hank Sedin is known for taking a lot of penalties, part of this is due to his heavy ice time, so when we look at penalties per-60, he’s only the fourth worst offender on the team (looking at you, Aaron Volpatti). We also have the ability to see penalties drawn per-60, which is fun and useful.
Game Situation Filters – some sites let you filter for power play and shorthanded situations as well. I’ll dive into this more in the future, but a lot of my advanced stats analysis will focus on 5-on-5 play. Odd-man situations are more complex and also operate in a much smaller sample size, making it difficult to trust the numbers as much as even-strength numbers.
Behind the Net is probably the first stop for any stat-happy hockey fan. This FAQ they provided way back when on Arctic Ice Hockey is a very good starting place for your reading, if you’re more interested.
CORSI – CORSI is basically a plus-minus rating based on shots instead of goals. The basic premise is that the goal of a line is to out-play the other line, and shots can be a good proxy for that. Shots here include shots that missed the net or were blocked, since we don’t care about evaluating the goalies. Missed nets and blocked shots help us get a better idea of puck dominance and zone control. When the Sedins were on the ice, for example, the Canuckstook way more shots than they allowed, and thus the Sedins have very high CORSI scores.
CORSI REL – “Relative” CORSI attempts to gauge a player’s value by comparing the team’s CORSI when he’s on the ice to the team’s CORSI when he’s off the ice. That is, even thoughMarc-Andre Gragnani doesn’t have the strong CORSI that some others have, the team had a poor CORSI when he was off the ice, which shows that the team did worse without him. Thus, he has a strong CORSI-REL.
CORSI QOC (Quality of Competition) – This is the CORSI score of the players that a player played against. So, in our example above, Harold Hustle plays against very good players who are likely to have a strong CORSI, thus his CORSI QOC would probably be high. Last year, the Canucks had negative CORSI QOC for almost every player (Sammy Pahlsson andMason Raymond were the exceptions), showing they had a somewhat favourable strength of schedule. It also shows that Pahlsson and Raymond were tasked with matching up against very tough opposition, which is important to note.
CORIS REL QOC – This is the CORSI REL score of the players the player played against. On one hand, this gives us a better idea of whether players were matching up against a team’s “top guys.” On the other, CORSI REL of an opponent doesn’t really matter much for player analysis, since it is only comparing an opposing player to his teammates. I don’t get a lot of value out of this stat, but thought I should highlight it anyway.
CORSI REL QOT This measures the CORSI REL of a player’s accompanying teammates when he’s on the ice and is pretty straightforward given the earlier explanations. Alex Burrows has a very high CORSI REL QOT because he played with the Sedins a lot. Dale Weise played on the fourth line and didn’t have All-Stars as linemates, so his CORSI REL QOT is low.
Zone Start and Zone Finish – These percentage stats show the frequency with which players started and ended their shifts in the opposing team’s zone. The Sedins, for example, started their shifts in the offensive zone a ridiculous 79% of the time, while Manny Malhotra was tasked with starting in his own zone an insane 87% of the time. Further to this, Malhotra finished his shift in the opposing zone 41% of the time – it doesn’t take a math major to figure out that moving the play from your own zone to the other team’s zone that often is very valuable.
PDO – PDO sums the shooting percentage of the team and the save percentage of the team when a player is on the ice. Basically, the metric assumes that all players and teams will eventually shoot a certain percentage, and any fluctuations around that percentage are at least partially subject to luck/randomness (think BABIP in baseball, if you’re familiar). While I disagree with the premise in general (some players will shoot a higher percentage or a lower percentage, some teams have bad goalies, etc), the metric regresses these stats to a long-term mean anyway. The stat is scaled to 1000, so someone with a PDO of 1050 (say,Byron Bitz) has been fortunate, while someone with a PDO of 975 (David Booth) has been somewhat unfortunate.
Important Note: FENWICK stats are basically the same as CORSI but exclude blocked shots. While they are shown to be more predictive of future performance at the team level, it’s not worth explaining them further. Basically, people are free to choose whether they want to include blocked shots (CORSI) or not (Fenwick). I choose to use CORSI because it is more accessible via Behind The Net and better for individual player analysis, but you can certainly craft an argument that, at least at the team level, blocked shots should be excluded since they’re a demonstrable skill.
This site is behind a pay-wall but has some good content available for free, primarily GVT.
GVT – Goals Versus Threshold is similar to “Wins Above Replacement” or “WARP” or “VORP” in other sports, as it aims to measure a player’s worth relative to fringe NHLers. It’s measured in goals and is thus easy to understand – losing Ryan Kesler would cost the Canucks 8.4 goals if they replaced him with a fringe NHL player. It can also be broken down into offensive and defensive components to show where a player derives his value (Kesler, for example, provides value at both ends).
This site lets you look upl CORSI, FENWICK, and their own ratings, by team, and also based on game situation. I have no idea how their own ratings hold up compared to CORSI/Fenwick, but I do know, thanks to their search tools, that Marc-Andre Gragnani and Andrew Alberts were the most valuable (meaning the best CORSI in this case) Canucks at even strength down by a goal while Chris Higgins derived most of his value when the team was ahead by two or more goals. It doesn’t tell us a lot, but it’s a fun little addition to help you see the value in these stats.
Moving back to our Hustle/Dangles example…
Hustle would be likely to have a weak CORSI REL QOT but a strong CORSI QOC. Given his faceoff prowess, his zone percentages would probably also be favourable, similar to a Malhotra-type in terms of moving the play from one end to the other. Without crunching the numbers, Hustle is exactly the type of player who may be undervalued by “hockey card” stats.
I didn’t say much negative about Dangles, and he was mostly used as a comparison point for Harold Hustle. However, we know his CORSI REL QOT would be high given his linemates, and his per-60 numbers would move closer to Hustle’s given the boost he gets from ice time. Dangles might very well be a superstar, I didn’t provide enough information to tell, but the gap between him and Hustle is definitely smaller than just points would suggest.
I’ll double back later in the weekend to use some of these stats to evaluate potential second-line centre options and further illustrate the stats.
Leave any questions in the comments, feel free to ask me to explain any other stats (or these ones in greater detail), and I’ll do my best. Also give me a follow, because I’ll be coming back every so often.